24.03 Liontrust Global Innovation Report - The Rise of AI 04.24 - Flipbook - Page 31
Much of the daily lifeblood of companies like ours relies on text –
emails, clinical reports, standard operating procedures that detail
manufacturing processes. A huge hurdle in getting new medicines
to patients is navigating the arduous clinical trial and regulatory
approval processes. Phase 3 trials can have 30,000 participants
– tracking everything that happens requires vast amounts of data.
Regulations aren’t designed to slow things down, they are there
to ensure that things are being done the right way. Accelerating
these processes by using computer programs that can understand
language is amazingly helpful.
One simplistic application is accelerating responses to questions
posed by regulators. These questions happen throughout the
development process, and come in the form of an email with
something like 100 questions. Each question requires its own entry,
and traditionally these must be manually entered into a database
system with metadata tagging it to the appropriate part of the
process – for example, this question is about manufacturing, this
is about clinical trials, and so on. AI and LLMs can automate this
by predicting metadata based on the question text, and pre-fill
everything for review and submission. This compresses an entire
day of work into five minutes, freeing up considerable time.
Deeper applications include expediting the research and response
processes. We are a platform company and most of what we
have done so far is vaccines, so we get the same types of
questions repeatedly. The raw material to answer these questions
is already in text form in our systems. But having someone find
and read a 500-page document to find one thing, and repeating
this for probably the 20th time because this is a question we
get regularly, is inefficient and tedious. This can be accelerated
dramatically with AI and LLMs, and this novel technology only
continues to improve.
These models could once only handle a limited amount of text at a
time – state-of-the-art used to be 100,000 tokens (word parts). But
we are seeing rapid progress such as Google’s most recent model
which can process 10 million tokens in a research context. This
increased capacity means you can chuck in all the information
and documents that you have collected as part of this regulatory
process and get to a first draft very quickly. Validation is still
needed, but this takes closer to 30 minutes compared to 10 hours
for manual draft creation. It’s not perfect, but this is a massive time
saving, and these models are only going to get better from here.
This is going to rapidly change every part of the development
process.
How does AI impact the competitive dynamics of the
industry? Will companies who have embraced AI
be in a better competitive position?
AI is one of the most powerful technologies that will
emerge in our lifetimes and will eventually be completely
pervasive. From our perspective, mRNA is a new technology. We
have ambitious pipelines and a relatively small workforce, so need
to be as efficient as possible, and this is why we’ve essentially run
towards AI/ML.
Many companies just like us have banned these tools. We’re on the
opposite end of the spectrum: not only do we encourage people
to use it, we also want to educate our people to be able to use
it effectively. We want to change the way they think about doing
their job. I supervise two software teams dedicated to building
capabilities so this can be embedded into everything we do.
How are you seeing AI advance in your business?
There’s not a part of the business that isn’t touched by
AI/ML. I’ve already touched on how we are using
it in drug design in areas like optimising our mRNA
sequences, protein sequences, and formulations,
and how we are using generative AI and LLMs for optimising
communications with regulators.
We are also using it in clinical trials for forecasting enrolment to
help us understand exactly how long a trial will take and how to
optimise advertising spend and targeted community outreach to
drive engagement. In manufacturing, we are using it extensively for
schedule optimisation. One example of where we apply this is for
our personalised cancer vaccines, where the turnaround from patient
biopsy/tumour sequencing to us producing a bespoke medicine
for them is incredibly short – around 45 days – so efficiency and
process optimisation is crucial when we do this at scale. We also
use it earlier in our research process when doing things like animal
studies. From a commercial perspective, as we continue to grow,
there are also plenty of opportunities to use AI/ML to optimise
areas like marketing. ML first got traction in e-commerce and digital
marketing spaces; this approach is applicable to our products too.
Internally, we deployed an application called mChat which is essentially
an internal ChatGPT, and we are hearing many stories of people
finishing tasks in 20 minutes instead of entire days. This transformation is
happening across every department and knowledge worker.
What has it taken as an organisation to fully
embrace AI? What do companies need to do to be
on the right side of this in terms of investment and
resource building?
A major thing is simply making AI available, not
banning it. Given all the capabilities this unlocks, it will change how
people use computers, so effective change management is needed.
At Moderna, we have 12 “mindsets” which are aligned with a
culture of curiosity-driven innovation, and this has helped our speed
of adoption.
Internal outreach is important. We have held townhalls, run workshops,
and set up a special interest group dubbed the Generative AI
Champions Team. These folks are constantly surprising us with novel
applications they can find, and act as advocates for generative AI
within their own business departments. Investing in and empowering
these people to demonstrate the value of AI and spread adoption
through word-of-mouth is really important. You reach this point of no
return where people can’t imagine going back to their jobs without AI.
But it takes effort and investing in your workforce to get there.
Companies will also need to hire for this capability. This is an
emerging skillset that hiring managers are going to need to pay
attention to really get the most out of these tools in their teams.
Do you see potential drug discovery use cases for AI
in modalities beyond mRNA?
In general, this is a tool that eventually everyone is
going to adopt in various ways. But there are some
applications that are unique to our modality and the
specific problems that we are trying to solve in mRNA and protein
sequencing and engineering. There’s no shortage of interesting
problems that we can apply it to. Our platform approach gives us
a good ROI on these types of early innovation investments, and that
has been really important for our success.
The rise of AI: Technology and Innovation Report - 31